cordmaur

dls = db.dataloaders(source=imgs_path, bs=4, num_workers=0)

dls.show_batch(figsize=(10,10))

cordmaur

ds = db.datasets(source=imgs_path)

idx = 1
img = ds.train[idx][0]
msk = ds.train[idx][1]

_, ax = plt.subplots(1, 2, figsize=(10, 5))
img.show(ctx=ax[0], bright=2.)
msk.show(ctx=ax[1])

cordmaur

def get_lbl_fn(img_fn: Path):
    lbl_path = img_fn.parent.parent/'labels'
    lbl_name = img_fn.name
    return (lbl_path/lbl_name)

db = DataBlock(blocks=(TransformBlock(type_tfms=partial(MSTensorImage.create, chnls_first=True)),
                       TransformBlock(type_tfms=[get_lbl_fn, partial(open_npy, cls=TensorMask)], 
                                      item_tfms=AddMaskCodes(codes=['clear', 'water', 'shadow'])),
                      ),
               get_items=partial(get_files, extensions='.npy'),

cordmaur

mask = TensorMask(open_npy(msk_path))
print(mask.shape)

_, ax = plt.subplots(1, 2, figsize=(10, 5))
img.show(bright=3., ctx=ax[0])
mask.show(ctx=ax[1])

cordmaur

img = MSTensorImage.create(img_path)
print(img)

_, ax = plt.subplots(1, 3, figsize=(12, 4))
img.show(bright=3., ctx=ax[0])
img.show(chnls=[2, 7, 10], ctx=ax[1])
img.show(chnls=[11], ctx=ax[2])

cordmaur

def open_npy(fn, chnls=None, cls=torch.Tensor):
    im = torch.from_numpy(np.load(str(fn))).type(torch.float32)
    if chnls is not None: im = im[chnls]
    return cls(im)

class MSTensorImage(TensorImage):
    
    def __init__(self, x, chnls_first=False):
        self.chnls_first = chnls_first

cordmaur

# Plotting a sample
_, ax = plt.subplots(1, 2, figsize=(10, 5))
ax[0].imshow(img.transpose((1, 2, 0))[..., [3, 2, 1]]*3.0)
ax[1].imshow(msk)

cordmaur

# Checking files
path = Path('./Data/')
imgs_path = path/'images'
lbls_path = path/'labels'

print(f'Checking number of files - images:{len([f for f in imgs_path.iterdir()])}\
      masks:{len([f for f in lbls_path.iterdir()])}')


# Checking file shapes 

cordmaur

def expand_block(in_channels, out_channels, kernel_size, padding):

    expand = nn.Sequential(
        torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=padding),
        torch.nn.BatchNorm2d(out_channels),
        torch.nn.ReLU(),
        torch.nn.Conv2d(out_channels, out_channels, kernel_size, stride=1, padding=padding),
        torch.nn.BatchNorm2d(out_channels),
        torch.nn.ReLU(),
        torch.nn.ConvTranspose2d(out_channels, out_channels, kernel_size=3, stride=2, padding=1, output_padding=1) 

cordmaur

    def __getitem__(self, idx):
        
        x = torch.tensor(self.open_as_array(idx, invert=self.pytorch, include_nir=True), dtype=torch.float32)
        y = torch.tensor(self.open_mask(idx, add_dims=False), dtype=torch.torch.int64)
        
        return x, y
    
    def open_as_pil(self, idx):
        
        arr = 256*self.open_as_array(idx)